Step 3.7 Flash (High): benchmark results
Provider: StepFun. Released 2026-05-29. Access: Open.
Aggregate Rating 1601 ± 19, rank #722 of 2170 rated models, from 24 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| SceneActBench - Layout | 77.5 | Task score (0-100): 100 3D-FRONT rooms from one view: place | 80 |
| SceneActBench - Dynamic | 57.9 | Task score (0-100): 10 low-poly Kenney scenes from a sampled | 50 |
| ObviousBench | 90.28 | Answer pass³ (%) | 44.5 |
| Context Arena | 35.33 | Average Score (%) | 31.6 |
| SuperCLUE-SWE - Overall | 40 | Score | 28.1 |
| PACT (Lechmazur) | 1500 | PACT Bilateral Rating | 28 |
| SceneActBench | 41.1 | Overall Score (%) | 20 |
| SceneActBench - Reconstruction | 8.6 | Task score (0-100): 100 3D-FRONT rooms from about 11 calibra | 20 |
| SuperCLUE-Law - Overall | 77.19 | Score | 20 |
| Multi-turn Debate (Lechmazur) | 1373.8 | Bradley-Terry Rating | 13.3 |
| SuperCLUE General (July 2026) - Agentic Task Planning | 66.54 | Score | 11.8 |
| SQBench - L1 Atomic Capabilities | 27 | Strict Pass@1 (%; the 100 L1 atomic-capability tasks on SQBe | 11.5 |
Interactive version: theaggregate.ai/model?slug=step-3-7-flash-high · How It Works · Data refreshed daily, snapshot 2026-11-05.